Q&A: Nnaisense talks AI and energy efficiency for data centers

The applications for artificial intelligence are growing by the day. Using AI to reduce energy consumption is one of the more important applications of this technology, and the addressable market at data centers around the world is significant.

We recently spoke with leading AI applications research company Nnaisense executive vice president Ralf Haller to learn more.

What is driving data centers to improve the efficiency of their energy consumption? Costs? Other?

Energy efficiency is now a central issue for data centers, who often overlooked this issue in the past. For the most part, we can chalk this up to two things: costs and climate change.

With respect to costs, the use of data is reflected in the energy costs of running and maintaining data centers. Seeing as the overwhelming majority of IP traffic travels through data centers, ordering a pair of shoes online or watching your favourite Netflix series causes these systems to use energy. These energy bills are astronomical, especially if the data center is not optimised.

One of the incumbent telecoms here in Switzerland, where Nnaisense is headquartered, racks up about $22.8M in energy costs every year. Energy efficiency, in this context, equates to lower costs and with the technology out there to make this possible, optimising energy usage is a real focus for data center operators.

In terms of climate change, the global mission to reach net zero is having a large impact on data centers. Corporate images are at stake, along with a responsibility to help save our planet from irreversible damage. Because data centers use a huge amount of electricity to run their equipment and keep it cool, they present a key challenge to meeting climate change goals. There is a growing focus on the need to improve their energy efficiency for this very reason.

How can AI help with this?

AI can optimise data centers as entire systems. Currently, each HVAC device, often doing nothing at all, runs at full, half or no speed just to cool nearby IT loads with no 'knowledge' of the whole data center system.

AI utilises sensor input data, such as temperature at the rack or room level, air pressure data, ambient weather data and IT consumption data to predict the optimal settings for the cooling liquid flow of each HVAC device. The important thing is that the data center is managed as a whole system with constraints for each device.

Digital twin technology is a great use-case of AI applied to data centers. A virtual clone of the entire operating system is created using real-time data, enabling the user to achieve optimisation through experimentation, which is then applied to the actual data center.

Where is the demand coming from both geographically and vertically?

Currently, AI installations can be found in data centers, primarily in North America and China. Europe is following suit, with more frequent reports of the successful use of AI in this domain. Europe may actually set the trend in this area going forward.

As signatories to the Climate Neutral Data Center Pact Self-Regulatory Initiative Policy Pledge - a part of the European Green Deal - dozens of top cloud service providers, from Google and Amazon on down, are committing themselves to completely power their installations with carbon-free energy by the end of 2030.

Can you point to any recent examples of successful projects in this direction?

Microsoft and Google Deepmind are two examples of successful AI deployment to optimise data efficiency. Another good example is a data center energy management solution installed by Huawei in its Linyi Big Data Center.

According to the company, its AI-based solution reduced UPS energy consumption alone by 40%, "enjoying a system efficiency of 97% and a module efficiency of 97.5%." In addition, it implemented a special cooling system that uses deep learning to correlate IT loads, environmental variables and equipment capabilities.

AI is used to chill water in the cooling system to achieve maximum energy efficiency, and data gathered from sensors "executes the instructions issued by the algorithms, including adjusting the amount of operating equipment; adjusting target values of control loops like rotational speed, power, temperature and pressure difference; and switching cooling mode.

Huawei uses its AI-based cooling and energy efficiency systems for its internal operations. Still, data center companies worldwide can partner with AI technology organisations to achieve similar results. The AI technology to reduce the carbon footprint of data centers exists right now. Regulators are likely to require they do so anyway, and their customers - both business and consumers - are demanding it.

What advice do you have to share with data centers for planning and managing such a project?

To successfully implement AI-driven optimisation technology, it is essential to work with a highly experienced external AI team who specialises in this field. Deep AI deployment knowledge is integral to a successful project.

Whilst AI can automate and optimise many industrial processes, including data centers, expert oversight is a fundamental prerequisite to navigating the installation process, troubleshooting issues as and when they arise and streamlining maintenance. This is why only big tech firms have implemented AI into their operations to hit their optimisation objectives to date.